Journal "Software Engineering"
a journal on theoretical and applied science and technology
ISSN 2220-3397
Issue N2 2017 year
Approaches to the problem of the development of rational (seeking optimal) planning techniques (drawing up the framework, the skeleton program in terms of parallelism) to perform the tasks of parallel computing systems with ho-mogeneous and heterogeneous field calculators. The informational graphs are used as a formalization of the algorithm. The possible criteria and parameters of the optimization planning methods with the use of stacked parallel form (SPF) information graph algorithm, or without it are described. Two basic strategies are formulated — preserving run time (without increasing the height of the original SPF) or an increase in the height of SPF. The first one is implemented on the basis of statements on location variability tiers SPF (while retaining the same information dependencies as in the original graph), the second one — by adding tiers in SPF and the transfer of the operators "top-down" on the written stage. The data on the developed for the implementation of such methods software system (the program stand) are provided. To implement methods (strategies) of rational development of parallel programs, the built-in high-level Lua scripting language is used. The confirmation of its effectiviness in this capacity is provided. The information on the set of API calls of the system is provided. A criterion of computational complexity of parallel programs execution plan building procedures is suggested. We give qualitative (in the form of strip-chart) and quantitative results of the application of some of the proposed strategies for building a rational planning of parallel programs with respect to the common algorithms of data processing.